液晶与显示2025，Vol.40Issue(10)：1509-1519,11.DOI:10.37188/CJLCD.2025-0138
基于深度学习的高分辨率夜光与光学遥感影像配准
Registration of high-resolution nighttime light and optical remote sensing images based on deep learning
摘要
Abstract
As heterogenous remote sensing images,high-resolution nighttime light and optical remote sensing images exhibit significant differences,making it impossible to achieve automatic registration based on traditional image registration algorithms.Currently,registration primarily relies on manual methods using ArcGIS to manually pinpoint points for registration.Therefore,we propose a deep learning-based automatic registration framework for high-resolution nighttime light and optical remote sensing images.Firstly,the binary road networks of nighttime light and optical remote sensing images are extracted and downsampled,and the sum of absolute differences(SAD)algorithm is used for coarse matching between the two types of images.Secondly,the YOLOv8 object detection model is utilized to extract the center points of road network intersections in nighttime light and optical remote sensing images as control points.The Euclidean distance and random sample consensus(RANSAC)algorithms are then employed to match and filter out corresponding control points.Finally,the least squares method is adopted to solve for the affine transformation matrix,achieving precise registration of high-resolution nighttime light and optical remote sensing images.The effectiveness of the proposed method is validated using Jilin-1 nighttime light remote sensing images with a resolution of 0.92 m and optical remote sensing images with a resolution of 0.75 m.The experimental results demonstrate that the proposed method can achieve automatic registration of high-resolution nighttime light and optical remote sensing images.The root mean square error(RMSE)after registration for test data from the built-up areas of Chengdu and Changchun are 3.29 m and 3.36 m,respectively,indicating high registration accuracy.关键词
夜光遥感影像/光学遥感影像/深度学习/路网/自动配准Key words
nighttime light remote sensing image/optical remote sensing image/deep learning/road network/automatic registration分类
信息技术与安全科学引用本文复制引用
孙鹏韬,李建荣,王志乾,于树海..基于深度学习的高分辨率夜光与光学遥感影像配准[J].液晶与显示,2025,40(10):1509-1519,11.基金项目
吉林省科技发展计划(No.20220201096GX)Supported by Jilin Province Science and Technology Development Plan(No.20220201096GX) （No.20220201096GX）
